ComPDF AI User Guide

1. ComPDF AI Overview

ComPDF AI provides ready-to-use solutions for intelligent document extraction, enterprise knowledge bases, and document splitting. It converts unstructured documents into structured data and supports automated business workflows.

Module	Description	Typical use cases
Intelligent document extraction	Extracts fields from invoices, contracts, policies, IDs, and other documents using semantic understanding and key-value pair technology.	ERP entry, invoice review, contract checks
Intelligent knowledge base	Builds a private knowledge base with AI-assisted search, answers, and source citations.	Support, training, compliance, internal search
Document splitting	Splits large PDF files into smaller documents using flexible page rules.	Archiving, distribution, preprocessing

2. Account and Team Management

After signing in, use the left navigation or home shortcuts to open a product module.

2.1 Sign in

1. Open the ComPDF AI sign-in page.
2. Enter your username or email address and password.
3. Accept the Terms of Service and Privacy Policy.
4. Click Sign in.

You may select Keep me signed in for faster access next time. Contact a super administrator if you forget your password.

2.2 Personal Center

Select your account icon in the upper-right corner and choose Personal Center. You can update your display name, avatar, and language. The language switcher is also available in the lower-left corner.

2.3 User Management

Only administrators can access user management. Select the account icon, choose Admin Console, and open User Management.

To add a member, choose Invite or Bulk invite, then enter registered member email addresses or upload a list. Administrators can view members, assign roles and permissions, reset passwords, and remove members.

3. Home Dashboard

The home dashboard summarizes tasks in the current user's organization scope and provides a quick Upload files entry point.

3.1 Data Overview

The overview cards show:

• New today: WorkFlow files uploaded today.
• Pending review: Completed extraction results that have not been confirmed.
• Failed: Parsing, classification, or extraction failures.
• Exceptions: Field-level exceptions reported by extraction validation.

The dashboard refreshes when the page loads. A zero value is valid and the card can still be opened to view an empty list.

3.2 Quick Actions

The main quick action is Upload files. It opens the upload dialog, where you choose files and a processing mode; after submission, the system opens the History Files list to show progress.

4. Intelligent Document Extraction (WorkFlow)

4.1 Overview

WorkFlow converts documents into structured JSON, Excel, or CSV data. The normal process is:

Upload -> classify or select a template -> extract -> review -> confirm or export

Supported inputs include PDF, Office documents, images, TXT, and CSV. A single file can be up to 100 MB and a batch can contain up to 32 files.

4.2 Template Management

Templates define the fields and extraction rules for a document type. Open Intelligent Document Extraction -> Template Settings.

4.2.1 Create and manage templates

Click New template, upload one or more sample files, enter a template name, configure fields, and save. Existing templates can be edited or deleted. The template ID can be copied for API use.

4.2.2 Basic field configuration

For each field, configure the following information:

Setting	Description	Required
Field name	The displayed name used to identify extracted content.	Yes
Prompt	An instruction that guides extraction. For example: “Extract the total amount without a currency symbol.”	No

Prompts can significantly improve extraction accuracy. They guide the model to locate and extract the intended information, especially when formats vary or fields are easily confused.

4.2.3 Output configuration

Output configuration controls the exported field name, type conversion, formatting, and validation. Supported output types include number, text, date, and enumeration. You can test a configuration with sample values before saving.

Path: in the template editor, click More settings at the end of a field row.

Output configuration also supports an export field name (letters, numbers, and underscores; maximum 64 characters; unique within the template), conversion-failure handling, and validation. Supported validations are required-value checks, format checks for number/text fields, and comparison checks for number/date fields. Use Run Test with a sample raw value to verify a configuration without saving it.

The complete conversion options are:

Output type	Rules
Number	Remove currency symbols such as ￥, $, €, and £; remove thousands separators; retain 0-6 decimal places; omit trailing zeros; and convert units such as thousand, ten thousand, million, or hundred million.
Text	Remove leading/trailing or all spaces and line breaks; convert case; replace text in sequence; and extract with regular expressions, including common phone, email, and ID patterns.
Date	Recognize common date formats, including date-time values, and normalize output to formats such as YYYY-MM-DD, YYYY/MM/DD, DD/MM/YYYY, or YYYY年MM月DD日.
Enumeration	Define a standard value list, map raw extracted values to standard values, and apply mappings in their configured order.

For failed conversion, choose No processing, Output empty, Output default value, or Mark exception. Validation includes required-value validation for all types, format validation for number/text, and comparison validation for number/date values.

4.2.4 AI-assisted configuration

After uploading a sample, click AI configuration to identify fields and generate an initial configuration. Review and edit the generated result before saving.

4.2.5 Default templates

ComPDF AI includes default templates for common invoices, orders, contracts, and other business documents. Default templates can be used directly; create a custom template when additional fields or rules are required.

4.3 File Processing

4.3.1 Upload files and select a mode

Click Upload files from the WorkFlow page or the home dashboard. Drag files into the upload dialog or select them from the local file system, then choose a processing mode.

1. Click an upload entry point to open the upload dialog.
2. Drag files into the dialog or click to select them.
3. Select one or more files.
4. Select a processing mode.
5. Click Confirm. The system starts upload automatically.

• LayoutExtract: Parses the document layout first and then extracts fields. It supports large documents, cross-page tables, and result traceability.
• VisualExtract: Uses a visual model to extract each page. It is suitable for regular or single-page documents and handwritten text.

The processing mode cannot be changed after upload; upload the file again to use another mode.

After upload, the selected mode creates an asynchronous processing task. Asset quota is reserved from estimated page count, settled against actual pages after success, and released when a task fails or is cancelled. One failed file does not interrupt the remaining files in a batch.

4.3.2 History file list

The history list shows file name, template, mode, processing status, and review status.

Field	Description
File name	Name of the uploaded file.
Template name	Template associated with the file.
Processing mode	LayoutExtract or VisualExtract.
Processing status	Pending parsing, parsing, parsing completed, pending classification, classifying, pending extraction, extracting, extraction completed, parsing failed, classification failed, or extraction failed.
Review status	Pending or confirmed. Displayed after extraction completes.

Available actions depend on the processing status: open details, remove a file, retry a failed file using its original mode, change classification and re-extract, or re-run extraction for a completed file.

4.3.3 File details

Open a file from the history list to view the source document and extraction result.

Area	Content
Left sidebar	Files under the current template; supports file-name search and switching files.
Center	PDF preview with zoom, page navigation, and bidirectional anchor positioning.
Right	Result tabs for layout and extraction results.

Layout results are available only for LayoutExtract. Extraction results are available for both modes. Before confirmation, fields can be corrected and results can be downloaded as JSON, Excel, or CSV.

Action	Condition	Result
Confirm	Review status is Pending and the user clicks Confirm.	Review status becomes Confirmed, results become read-only, and confirmer/time are recorded.
Revoke confirmation	Review status is Confirmed and the user clicks Revoke confirmation.	Enter a required reason of up to 200 characters; review returns to Pending and results become editable.

4.4 Results Dashboard

The results dashboard presents extraction results in a table and supports column configuration, search, template filters, and exception filters.

4.4.1 Review and confirmation

Compare the result with the source document and click Confirm when the result is correct. Confirmed results become read-only. Use Revoke confirmation and provide a reason to edit the result again.

4.4.2 Column configuration

Use column configuration to select and reorder the text fields displayed in the dashboard. Available fields are grouped by template and shown by export field name. Under All Files, fields with the same name across templates are merged; under a selected template, only that template's text fields are shown.

4.4.3 Exception markers

The service validates fields when extraction completes and when a result is saved. Invalid fields are shown with a red background, red border, and warning marker.

5. API Key and Webhook Management

Open the account menu, choose Admin Console, and open the API section. API Keys are used by external services; Webhooks deliver asynchronous events to a business endpoint.

5.1 API Key management

5.1.1 Create an API Key

1. Open API → API Keys.
2. Click Create API Key.
3. Enter a required key name of up to 50 characters so the calling system or business scenario can be identified.
4. Choose an expiration period: Never, two weeks, one month, three months, six months, or one year.
5. Confirm creation and copy the complete key immediately. The complete value is shown only once.

Use the key in API requests as follows:

Authorization: Bearer <YOUR_API_KEY>

The list only displays a masked key. Never place a complete key in browser code, public repositories, screenshots, or logs. Disable or delete a leaked key and create a replacement.

5.1.2 Manage API Keys

Use the status switch to enable or disable a key. An expired key cannot be enabled. Delete a key only after confirming the action; deletion is permanent. The list displays a masked key only; do not place plaintext keys in browser code, repositories, screenshots, or logs. Disable or delete a leaked key and create a replacement.

5.2 Webhook management

5.2.1 Add a Webhook

1. Open API → Webhooks and click Add Webhook.
2. Enter an HTTP or HTTPS endpoint. HTTPS is recommended for production.
3. Select one or more subscription events:
   • Batch started
   • Batch completed
   • Document processing started
   • Document processing succeeded
   • Document processing failed
   • Rule-triggered notification
4. Save the Webhook and copy its signing secret.

The receiver must verify the X-ComPDF-Signature header. A Webhook used by a reminder rule must be enabled and subscribed to the rule_triggered event. Failed deliveries are recorded and retried according to the service policy.

5.2.2 Manage Webhooks

Use the status switch to enable or disable a Webhook; the page displays the operation result. To delete a Webhook, click the delete icon and confirm. Deletion is permanent.

5.2.3 Signing secret and callback delivery

Every Webhook receives a unique signing secret of at least 32 characters. ComPDF AI sends a JSON HTTP POST request when a subscribed event occurs. The receiver must verify the X-ComPDF-Signature request header to confirm the callback source. Delivery timeout is 5 seconds; failed deliveries are retried every 5 minutes, up to three times. The most recent HTTP response code is retained for display.

6. Message Center

6.1 Messages

Open the message icon in the navigation bar or choose Message Center from the sidebar. The four top-level categories are All messages, System messages, System announcements, and User reminder rules. Rule messages are grouped by rule name.

Available actions include keyword search, mark all as read, open details, and delete one message.

6.2 Reminder settings

Click Settings in the Message Center to manage reminder rules. A rule contains a name, template data source, conditions, frequency, channels, recipients, title template, and content template.

Supported frequencies are daily, hourly, after extraction completes, and when a result is confirmed. Channels include in-app message, email, and Webhook. Organization members can receive in-app and email notifications; template field bindings are email-only and invalid email values are skipped.

The rule list supports enable, disable, edit, copy, and delete. A copied rule duplicates its configuration and starts disabled.

6.2.1 Rule list

Each card shows the rule name, source template, condition summary, delivery channels, and actions to enable, disable, edit, copy, or delete the rule.

6.2.2 Create or edit a rule

The condition selector provides the system fields Processing Status and Review Status, as well as text fields from the selected template. The notification template provides variables such as {file_name}, {template_name}, {detail_url}, and the selected template fields.

Configuration item	Requirement
Rule name	Required; maximum 50 characters; unique in an organization.
Data source	Select a template associated with the current organization.
Conditions	At least one complete condition: field, operator, and comparison value. Add or remove rows while retaining at least one; combine rows with AND or OR.
Frequency	Daily, hourly, after extraction completes, or when a result is confirmed.
Delivery channels	Select at least one of in-app message, email, and Webhook.
Members/departments	Required where member/department recipients are used; members must belong to the organization. Department membership changes are applied automatically.
Template fields	Email only. Values are validated as email addresses at delivery time; invalid or empty values are skipped and logged. Multiple comma-separated addresses are split.
Webhooks	Required when the Webhook channel is selected. Select active Webhooks subscribed to the rule-triggered event.
Title	Required; maximum 100 characters; supports variables.
Content	Required; maximum 500 characters; supports variables.

The same address from member, department, direct, or field sources is delivered once. A Webhook that is later disabled or deleted is skipped during rule execution; other channels continue normally. Failed mail and Webhook delivery does not interrupt the rule evaluation flow.

7. Intelligent Knowledge Base

7.1 Create and manage a knowledge base

Open Intelligent Knowledge Base, click Create knowledge base, and enter a name. Open a knowledge base to configure it or delete it from the management list.

7.2 Knowledge base configuration

Configure the name, cover, description, permissions, and chunking method.

7.2.1 Permissions

Add registered team members and assign View or Manage permission. Viewers can ask questions; managers can configure the knowledge base and manage its dataset.

7.2.2 Chunking

Choose a chunking method that matches the source material. Available methods include General, Q&A, Resume, Manual, Table, Paper, Book, Legal, Presentation, and Whole document.

7.3 Knowledge base dataset

7.3.1 Upload and parse files

Open the Dataset tab and click Upload files. You can start parsing automatically or upload first and parse manually. Supported formats include Office files, PDF, TXT, images, CSV, and JSON. You can also upload a web URL and choose a crawl depth.

7.3.2 Edit the dataset

The dataset list shows file name, format, update time, chunking method, chunk count, enabled state, and parsing state. Files can be renamed, downloaded, deleted, re-parsed, enabled, or disabled. Select multiple files for batch operations, or open a file to edit individual chunks.

You can perform batch operations on uploaded files: after selecting multiple files, click the Batch button shown below to batch enable, disable, parse, cancel parsing, or delete them.

Click a file name to view the parsed segments of that file; you can enable, disable, or delete the current parsed segments. Click the button shown below and enter the parsed segment, keywords, and question to create a new parsed segment.

7.4 Knowledge base Q&A

7.4.1 Create an AI assistant

Create an assistant with a name, description, avatar, empty-answer text, and welcome message. Associate one or more knowledge bases and choose whether citations are displayed.

7.4.2 Ask questions

After creating an assistant, enter a question in the Q&A page. The assistant answers using the associated knowledge bases and can show source citations.

8. Document Splitting

Open Document Splitting from the left navigation and upload PDF files. Select a file and choose one of the following rules:

• Split by page count.
• Split into a specified number of documents.
• Split by page range, odd pages, or even pages.

You can customize labels and separators in output file names. Click Split and download the generated files when processing completes.